Representing object shapes using radial basis function support vector machine classification

Abstract

A shape of an object is represented by a set of points inside and outside the shape. A decision function is learned from the set of points an object. Feature points in the set of points are selected using the decision function, or a gradient of the decision function, and then a local descriptor is determined for each feature point.

Description

RELATED APPLICATION[0001]This U.S. patent application is related to U.S. patent application Ser. No. 13 / 026,382, filed Feb. 14, 2011, co-filed herewith, and incorporated by reference.FIELD OF THE INVENTION[0002]This invention relates generally to representing shapes of objects, and more particularly to representing the shapes using radial basis function (RBF) support vector machine (SVM) classification.BACKGROUND OF THE INVENTION[0003]Shape Representation[0004]Shape of an object represents the geometrical information that is independent of the transformational (scaling, rotation, articulation, etc) effects. Understanding shape is essential in major computer vision applications from design and inspection in industrial manufacturing to content based retrieval in visual data search engines to surgical planning and monitoring of changes in medical systems to recognition of people and their actions in video surveillance to analysis of compositions in architecture and archeology.[0005]Rec...

AI Technical Summary

Benefits of technology

[0031]In contrast with conventional method, we consider a shape to be an analytic function of a decision boundary of the hyperplane of the SVM. This empowers the SVS representation with many advantages.

[0032]First, the representation uses a sparse subset of feature points determined by the support vectors, which significantly improves the discriminative power against noise, fragmentation and other artifacts in data.

[0033]Second, the use of RBF kernels provides affine invariant features, and allows any shape to be represented accurately regardless of its complexity.

[0034]Third, the SVS representation defines a new set of local descriptors that are determined from a highly stable decision function gradient. Therefore the SVS is robust against articulation.

[0035]Fourth, the representation is compact and easily extends to higher dimensions.

Problems solved by technology

The drawback of a purely global approach is the exclusion of articulation and the sensitivity to occlusion.

The drawback of a purely part-based approach is that a consistent partitioning is generally not possible in the face of numerous combinations of possibilities and object shape variations.

Besides, segmentation itself is ill-posed, except under controlled environments or for restricted application domains.

While shock graphs benefit from the skeleton's robustness to articulation they suffer from boundary noise.

While this method is simple and effective, it requires additional computation to classify testing data.

Images